xref: /dpdk/drivers/net/pcap/pcap_ethdev.c (revision bbbe38a6d59ccdda25917712701e629d0b10af6f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation.
3  * Copyright(c) 2014 6WIND S.A.
4  * All rights reserved.
5  */
6 
7 #include <time.h>
8 
9 #include <pcap.h>
10 
11 #include <rte_cycles.h>
12 #include <ethdev_driver.h>
13 #include <ethdev_vdev.h>
14 #include <rte_kvargs.h>
15 #include <rte_malloc.h>
16 #include <rte_mbuf.h>
17 #include <rte_mbuf_dyn.h>
18 #include <rte_bus_vdev.h>
19 #include <rte_os_shim.h>
20 
21 #include "pcap_osdep.h"
22 
23 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
24 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
25 #define RTE_ETH_PCAP_PROMISC 1
26 #define RTE_ETH_PCAP_TIMEOUT -1
27 
28 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
29 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
30 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
31 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
32 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
33 #define ETH_PCAP_IFACE_ARG    "iface"
34 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
35 #define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
36 
37 #define ETH_PCAP_ARG_MAXLEN	64
38 
39 #define RTE_PMD_PCAP_MAX_QUEUES 16
40 
41 static char errbuf[PCAP_ERRBUF_SIZE];
42 static struct timespec start_time;
43 static uint64_t start_cycles;
44 static uint64_t hz;
45 static uint8_t iface_idx;
46 
47 static uint64_t timestamp_rx_dynflag;
48 static int timestamp_dynfield_offset = -1;
49 
50 struct queue_stat {
51 	volatile unsigned long pkts;
52 	volatile unsigned long bytes;
53 	volatile unsigned long err_pkts;
54 };
55 
56 struct queue_missed_stat {
57 	/* last value retrieved from pcap */
58 	unsigned int pcap;
59 	/* stores values lost by pcap stop or rollover */
60 	unsigned long mnemonic;
61 	/* value on last reset */
62 	unsigned long reset;
63 };
64 
65 struct pcap_rx_queue {
66 	uint16_t port_id;
67 	uint16_t queue_id;
68 	struct rte_mempool *mb_pool;
69 	struct queue_stat rx_stat;
70 	struct queue_missed_stat missed_stat;
71 	char name[PATH_MAX];
72 	char type[ETH_PCAP_ARG_MAXLEN];
73 
74 	/* Contains pre-generated packets to be looped through */
75 	struct rte_ring *pkts;
76 };
77 
78 struct pcap_tx_queue {
79 	uint16_t port_id;
80 	uint16_t queue_id;
81 	struct queue_stat tx_stat;
82 	char name[PATH_MAX];
83 	char type[ETH_PCAP_ARG_MAXLEN];
84 };
85 
86 struct pmd_internals {
87 	struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
88 	struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
89 	char devargs[ETH_PCAP_ARG_MAXLEN];
90 	struct rte_ether_addr eth_addr;
91 	int if_index;
92 	int single_iface;
93 	int phy_mac;
94 	unsigned int infinite_rx;
95 };
96 
97 struct pmd_process_private {
98 	pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
99 	pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
100 	pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
101 };
102 
103 struct pmd_devargs {
104 	unsigned int num_of_queue;
105 	struct devargs_queue {
106 		pcap_dumper_t *dumper;
107 		pcap_t *pcap;
108 		const char *name;
109 		const char *type;
110 	} queue[RTE_PMD_PCAP_MAX_QUEUES];
111 	int phy_mac;
112 };
113 
114 struct pmd_devargs_all {
115 	struct pmd_devargs rx_queues;
116 	struct pmd_devargs tx_queues;
117 	int single_iface;
118 	unsigned int is_tx_pcap;
119 	unsigned int is_tx_iface;
120 	unsigned int is_rx_pcap;
121 	unsigned int is_rx_iface;
122 	unsigned int infinite_rx;
123 };
124 
125 static const char *valid_arguments[] = {
126 	ETH_PCAP_RX_PCAP_ARG,
127 	ETH_PCAP_TX_PCAP_ARG,
128 	ETH_PCAP_RX_IFACE_ARG,
129 	ETH_PCAP_RX_IFACE_IN_ARG,
130 	ETH_PCAP_TX_IFACE_ARG,
131 	ETH_PCAP_IFACE_ARG,
132 	ETH_PCAP_PHY_MAC_ARG,
133 	ETH_PCAP_INFINITE_RX_ARG,
134 	NULL
135 };
136 
137 static struct rte_eth_link pmd_link = {
138 		.link_speed = ETH_SPEED_NUM_10G,
139 		.link_duplex = ETH_LINK_FULL_DUPLEX,
140 		.link_status = ETH_LINK_DOWN,
141 		.link_autoneg = ETH_LINK_FIXED,
142 };
143 
144 RTE_LOG_REGISTER_DEFAULT(eth_pcap_logtype, NOTICE);
145 
146 static struct queue_missed_stat*
147 queue_missed_stat_update(struct rte_eth_dev *dev, unsigned int qid)
148 {
149 	struct pmd_internals *internals = dev->data->dev_private;
150 	struct queue_missed_stat *missed_stat =
151 			&internals->rx_queue[qid].missed_stat;
152 	const struct pmd_process_private *pp = dev->process_private;
153 	pcap_t *pcap = pp->rx_pcap[qid];
154 	struct pcap_stat stat;
155 
156 	if (!pcap || (pcap_stats(pcap, &stat) != 0))
157 		return missed_stat;
158 
159 	/* rollover check - best effort fixup assuming single rollover */
160 	if (stat.ps_drop < missed_stat->pcap)
161 		missed_stat->mnemonic += UINT_MAX;
162 	missed_stat->pcap = stat.ps_drop;
163 
164 	return missed_stat;
165 }
166 
167 static void
168 queue_missed_stat_on_stop_update(struct rte_eth_dev *dev, unsigned int qid)
169 {
170 	struct queue_missed_stat *missed_stat =
171 			queue_missed_stat_update(dev, qid);
172 
173 	missed_stat->mnemonic += missed_stat->pcap;
174 	missed_stat->pcap = 0;
175 }
176 
177 static void
178 queue_missed_stat_reset(struct rte_eth_dev *dev, unsigned int qid)
179 {
180 	struct queue_missed_stat *missed_stat =
181 			queue_missed_stat_update(dev, qid);
182 
183 	missed_stat->reset = missed_stat->pcap;
184 	missed_stat->mnemonic = 0;
185 }
186 
187 static unsigned long
188 queue_missed_stat_get(struct rte_eth_dev *dev, unsigned int qid)
189 {
190 	const struct queue_missed_stat *missed_stat =
191 			queue_missed_stat_update(dev, qid);
192 
193 	return missed_stat->pcap + missed_stat->mnemonic - missed_stat->reset;
194 }
195 
196 static int
197 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
198 		const u_char *data, uint16_t data_len)
199 {
200 	/* Copy the first segment. */
201 	uint16_t len = rte_pktmbuf_tailroom(mbuf);
202 	struct rte_mbuf *m = mbuf;
203 
204 	rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
205 	data_len -= len;
206 	data += len;
207 
208 	while (data_len > 0) {
209 		/* Allocate next mbuf and point to that. */
210 		m->next = rte_pktmbuf_alloc(mb_pool);
211 
212 		if (unlikely(!m->next))
213 			return -1;
214 
215 		m = m->next;
216 
217 		/* Headroom is not needed in chained mbufs. */
218 		rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
219 		m->pkt_len = 0;
220 		m->data_len = 0;
221 
222 		/* Copy next segment. */
223 		len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
224 		rte_memcpy(rte_pktmbuf_append(m, len), data, len);
225 
226 		mbuf->nb_segs++;
227 		data_len -= len;
228 		data += len;
229 	}
230 
231 	return mbuf->nb_segs;
232 }
233 
234 static uint16_t
235 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
236 {
237 	int i;
238 	struct pcap_rx_queue *pcap_q = queue;
239 	uint32_t rx_bytes = 0;
240 
241 	if (unlikely(nb_pkts == 0))
242 		return 0;
243 
244 	if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
245 		return 0;
246 
247 	for (i = 0; i < nb_pkts; i++) {
248 		struct rte_mbuf *pcap_buf;
249 		int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
250 		if (err)
251 			return i;
252 
253 		rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
254 				rte_pktmbuf_mtod(pcap_buf, void *),
255 				pcap_buf->data_len);
256 		bufs[i]->data_len = pcap_buf->data_len;
257 		bufs[i]->pkt_len = pcap_buf->pkt_len;
258 		bufs[i]->port = pcap_q->port_id;
259 		rx_bytes += pcap_buf->data_len;
260 
261 		/* Enqueue packet back on ring to allow infinite rx. */
262 		rte_ring_enqueue(pcap_q->pkts, pcap_buf);
263 	}
264 
265 	pcap_q->rx_stat.pkts += i;
266 	pcap_q->rx_stat.bytes += rx_bytes;
267 
268 	return i;
269 }
270 
271 static uint16_t
272 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
273 {
274 	unsigned int i;
275 	struct pcap_pkthdr header;
276 	struct pmd_process_private *pp;
277 	const u_char *packet;
278 	struct rte_mbuf *mbuf;
279 	struct pcap_rx_queue *pcap_q = queue;
280 	uint16_t num_rx = 0;
281 	uint32_t rx_bytes = 0;
282 	pcap_t *pcap;
283 
284 	pp = rte_eth_devices[pcap_q->port_id].process_private;
285 	pcap = pp->rx_pcap[pcap_q->queue_id];
286 
287 	if (unlikely(pcap == NULL || nb_pkts == 0))
288 		return 0;
289 
290 	/* Reads the given number of packets from the pcap file one by one
291 	 * and copies the packet data into a newly allocated mbuf to return.
292 	 */
293 	for (i = 0; i < nb_pkts; i++) {
294 		/* Get the next PCAP packet */
295 		packet = pcap_next(pcap, &header);
296 		if (unlikely(packet == NULL))
297 			break;
298 
299 		mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
300 		if (unlikely(mbuf == NULL))
301 			break;
302 
303 		if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
304 			/* pcap packet will fit in the mbuf, can copy it */
305 			rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
306 					header.caplen);
307 			mbuf->data_len = (uint16_t)header.caplen;
308 		} else {
309 			/* Try read jumbo frame into multi mbufs. */
310 			if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
311 						       mbuf,
312 						       packet,
313 						       header.caplen) == -1)) {
314 				rte_pktmbuf_free(mbuf);
315 				break;
316 			}
317 		}
318 
319 		mbuf->pkt_len = (uint16_t)header.caplen;
320 		*RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
321 			rte_mbuf_timestamp_t *) =
322 				(uint64_t)header.ts.tv_sec * 1000000 +
323 				header.ts.tv_usec;
324 		mbuf->ol_flags |= timestamp_rx_dynflag;
325 		mbuf->port = pcap_q->port_id;
326 		bufs[num_rx] = mbuf;
327 		num_rx++;
328 		rx_bytes += header.caplen;
329 	}
330 	pcap_q->rx_stat.pkts += num_rx;
331 	pcap_q->rx_stat.bytes += rx_bytes;
332 
333 	return num_rx;
334 }
335 
336 static uint16_t
337 eth_null_rx(void *queue __rte_unused,
338 		struct rte_mbuf **bufs __rte_unused,
339 		uint16_t nb_pkts __rte_unused)
340 {
341 	return 0;
342 }
343 
344 #define NSEC_PER_SEC	1000000000L
345 
346 /*
347  * This function stores nanoseconds in `tv_usec` field of `struct timeval`,
348  * because `ts` goes directly to nanosecond-precision dump.
349  */
350 static inline void
351 calculate_timestamp(struct timeval *ts) {
352 	uint64_t cycles;
353 	struct timespec cur_time;
354 
355 	cycles = rte_get_timer_cycles() - start_cycles;
356 	cur_time.tv_sec = cycles / hz;
357 	cur_time.tv_nsec = (cycles % hz) * NSEC_PER_SEC / hz;
358 
359 	ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
360 	ts->tv_usec = start_time.tv_nsec + cur_time.tv_nsec;
361 	if (ts->tv_usec >= NSEC_PER_SEC) {
362 		ts->tv_usec -= NSEC_PER_SEC;
363 		ts->tv_sec += 1;
364 	}
365 }
366 
367 /*
368  * Callback to handle writing packets to a pcap file.
369  */
370 static uint16_t
371 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
372 {
373 	unsigned int i;
374 	struct rte_mbuf *mbuf;
375 	struct pmd_process_private *pp;
376 	struct pcap_tx_queue *dumper_q = queue;
377 	uint16_t num_tx = 0;
378 	uint32_t tx_bytes = 0;
379 	struct pcap_pkthdr header;
380 	pcap_dumper_t *dumper;
381 	unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
382 	size_t len, caplen;
383 
384 	pp = rte_eth_devices[dumper_q->port_id].process_private;
385 	dumper = pp->tx_dumper[dumper_q->queue_id];
386 
387 	if (dumper == NULL || nb_pkts == 0)
388 		return 0;
389 
390 	/* writes the nb_pkts packets to the previously opened pcap file
391 	 * dumper */
392 	for (i = 0; i < nb_pkts; i++) {
393 		mbuf = bufs[i];
394 		len = caplen = rte_pktmbuf_pkt_len(mbuf);
395 		if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
396 				len > sizeof(temp_data))) {
397 			caplen = sizeof(temp_data);
398 		}
399 
400 		calculate_timestamp(&header.ts);
401 		header.len = len;
402 		header.caplen = caplen;
403 		/* rte_pktmbuf_read() returns a pointer to the data directly
404 		 * in the mbuf (when the mbuf is contiguous) or, otherwise,
405 		 * a pointer to temp_data after copying into it.
406 		 */
407 		pcap_dump((u_char *)dumper, &header,
408 			rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
409 
410 		num_tx++;
411 		tx_bytes += caplen;
412 		rte_pktmbuf_free(mbuf);
413 	}
414 
415 	/*
416 	 * Since there's no place to hook a callback when the forwarding
417 	 * process stops and to make sure the pcap file is actually written,
418 	 * we flush the pcap dumper within each burst.
419 	 */
420 	pcap_dump_flush(dumper);
421 	dumper_q->tx_stat.pkts += num_tx;
422 	dumper_q->tx_stat.bytes += tx_bytes;
423 	dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
424 
425 	return nb_pkts;
426 }
427 
428 /*
429  * Callback to handle dropping packets in the infinite rx case.
430  */
431 static uint16_t
432 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
433 {
434 	unsigned int i;
435 	uint32_t tx_bytes = 0;
436 	struct pcap_tx_queue *tx_queue = queue;
437 
438 	if (unlikely(nb_pkts == 0))
439 		return 0;
440 
441 	for (i = 0; i < nb_pkts; i++) {
442 		tx_bytes += bufs[i]->pkt_len;
443 		rte_pktmbuf_free(bufs[i]);
444 	}
445 
446 	tx_queue->tx_stat.pkts += nb_pkts;
447 	tx_queue->tx_stat.bytes += tx_bytes;
448 
449 	return i;
450 }
451 
452 /*
453  * Callback to handle sending packets through a real NIC.
454  */
455 static uint16_t
456 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
457 {
458 	unsigned int i;
459 	int ret;
460 	struct rte_mbuf *mbuf;
461 	struct pmd_process_private *pp;
462 	struct pcap_tx_queue *tx_queue = queue;
463 	uint16_t num_tx = 0;
464 	uint32_t tx_bytes = 0;
465 	pcap_t *pcap;
466 	unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
467 	size_t len;
468 
469 	pp = rte_eth_devices[tx_queue->port_id].process_private;
470 	pcap = pp->tx_pcap[tx_queue->queue_id];
471 
472 	if (unlikely(nb_pkts == 0 || pcap == NULL))
473 		return 0;
474 
475 	for (i = 0; i < nb_pkts; i++) {
476 		mbuf = bufs[i];
477 		len = rte_pktmbuf_pkt_len(mbuf);
478 		if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
479 				len > sizeof(temp_data))) {
480 			PMD_LOG(ERR,
481 				"Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
482 				len, sizeof(temp_data));
483 			rte_pktmbuf_free(mbuf);
484 			continue;
485 		}
486 
487 		/* rte_pktmbuf_read() returns a pointer to the data directly
488 		 * in the mbuf (when the mbuf is contiguous) or, otherwise,
489 		 * a pointer to temp_data after copying into it.
490 		 */
491 		ret = pcap_sendpacket(pcap,
492 			rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
493 		if (unlikely(ret != 0))
494 			break;
495 		num_tx++;
496 		tx_bytes += len;
497 		rte_pktmbuf_free(mbuf);
498 	}
499 
500 	tx_queue->tx_stat.pkts += num_tx;
501 	tx_queue->tx_stat.bytes += tx_bytes;
502 	tx_queue->tx_stat.err_pkts += i - num_tx;
503 
504 	return i;
505 }
506 
507 /*
508  * pcap_open_live wrapper function
509  */
510 static inline int
511 open_iface_live(const char *iface, pcap_t **pcap) {
512 	*pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
513 			RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
514 
515 	if (*pcap == NULL) {
516 		PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
517 		return -1;
518 	}
519 
520 	return 0;
521 }
522 
523 static int
524 open_single_iface(const char *iface, pcap_t **pcap)
525 {
526 	if (open_iface_live(iface, pcap) < 0) {
527 		PMD_LOG(ERR, "Couldn't open interface %s", iface);
528 		return -1;
529 	}
530 
531 	return 0;
532 }
533 
534 static int
535 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
536 {
537 	pcap_t *tx_pcap;
538 
539 	/*
540 	 * We need to create a dummy empty pcap_t to use it
541 	 * with pcap_dump_open(). We create big enough an Ethernet
542 	 * pcap holder.
543 	 */
544 	tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
545 			RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
546 	if (tx_pcap == NULL) {
547 		PMD_LOG(ERR, "Couldn't create dead pcap");
548 		return -1;
549 	}
550 
551 	/* The dumper is created using the previous pcap_t reference */
552 	*dumper = pcap_dump_open(tx_pcap, pcap_filename);
553 	if (*dumper == NULL) {
554 		pcap_close(tx_pcap);
555 		PMD_LOG(ERR, "Couldn't open %s for writing.",
556 			pcap_filename);
557 		return -1;
558 	}
559 
560 	pcap_close(tx_pcap);
561 	return 0;
562 }
563 
564 static int
565 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
566 {
567 	*pcap = pcap_open_offline(pcap_filename, errbuf);
568 	if (*pcap == NULL) {
569 		PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
570 			errbuf);
571 		return -1;
572 	}
573 
574 	return 0;
575 }
576 
577 static uint64_t
578 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
579 {
580 	const u_char *packet;
581 	struct pcap_pkthdr header;
582 	uint64_t pcap_pkt_count = 0;
583 
584 	while ((packet = pcap_next(*pcap, &header)))
585 		pcap_pkt_count++;
586 
587 	/* The pcap is reopened so it can be used as normal later. */
588 	pcap_close(*pcap);
589 	*pcap = NULL;
590 	open_single_rx_pcap(pcap_q->name, pcap);
591 
592 	return pcap_pkt_count;
593 }
594 
595 static int
596 eth_dev_start(struct rte_eth_dev *dev)
597 {
598 	unsigned int i;
599 	struct pmd_internals *internals = dev->data->dev_private;
600 	struct pmd_process_private *pp = dev->process_private;
601 	struct pcap_tx_queue *tx;
602 	struct pcap_rx_queue *rx;
603 
604 	/* Special iface case. Single pcap is open and shared between tx/rx. */
605 	if (internals->single_iface) {
606 		tx = &internals->tx_queue[0];
607 		rx = &internals->rx_queue[0];
608 
609 		if (!pp->tx_pcap[0] &&
610 			strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
611 			if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
612 				return -1;
613 			pp->rx_pcap[0] = pp->tx_pcap[0];
614 		}
615 
616 		goto status_up;
617 	}
618 
619 	/* If not open already, open tx pcaps/dumpers */
620 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
621 		tx = &internals->tx_queue[i];
622 
623 		if (!pp->tx_dumper[i] &&
624 				strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
625 			if (open_single_tx_pcap(tx->name,
626 				&pp->tx_dumper[i]) < 0)
627 				return -1;
628 		} else if (!pp->tx_pcap[i] &&
629 				strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
630 			if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
631 				return -1;
632 		}
633 	}
634 
635 	/* If not open already, open rx pcaps */
636 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
637 		rx = &internals->rx_queue[i];
638 
639 		if (pp->rx_pcap[i] != NULL)
640 			continue;
641 
642 		if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
643 			if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
644 				return -1;
645 		} else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
646 			if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
647 				return -1;
648 		}
649 	}
650 
651 status_up:
652 	for (i = 0; i < dev->data->nb_rx_queues; i++)
653 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
654 
655 	for (i = 0; i < dev->data->nb_tx_queues; i++)
656 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
657 
658 	dev->data->dev_link.link_status = ETH_LINK_UP;
659 
660 	return 0;
661 }
662 
663 /*
664  * This function gets called when the current port gets stopped.
665  * Is the only place for us to close all the tx streams dumpers.
666  * If not called the dumpers will be flushed within each tx burst.
667  */
668 static int
669 eth_dev_stop(struct rte_eth_dev *dev)
670 {
671 	unsigned int i;
672 	struct pmd_internals *internals = dev->data->dev_private;
673 	struct pmd_process_private *pp = dev->process_private;
674 
675 	/* Special iface case. Single pcap is open and shared between tx/rx. */
676 	if (internals->single_iface) {
677 		queue_missed_stat_on_stop_update(dev, 0);
678 		if (pp->tx_pcap[0] != NULL) {
679 			pcap_close(pp->tx_pcap[0]);
680 			pp->tx_pcap[0] = NULL;
681 			pp->rx_pcap[0] = NULL;
682 		}
683 		goto status_down;
684 	}
685 
686 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
687 		if (pp->tx_dumper[i] != NULL) {
688 			pcap_dump_close(pp->tx_dumper[i]);
689 			pp->tx_dumper[i] = NULL;
690 		}
691 
692 		if (pp->tx_pcap[i] != NULL) {
693 			pcap_close(pp->tx_pcap[i]);
694 			pp->tx_pcap[i] = NULL;
695 		}
696 	}
697 
698 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
699 		if (pp->rx_pcap[i] != NULL) {
700 			queue_missed_stat_on_stop_update(dev, i);
701 			pcap_close(pp->rx_pcap[i]);
702 			pp->rx_pcap[i] = NULL;
703 		}
704 	}
705 
706 status_down:
707 	for (i = 0; i < dev->data->nb_rx_queues; i++)
708 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
709 
710 	for (i = 0; i < dev->data->nb_tx_queues; i++)
711 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
712 
713 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
714 
715 	return 0;
716 }
717 
718 static int
719 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
720 {
721 	return 0;
722 }
723 
724 static int
725 eth_dev_info(struct rte_eth_dev *dev,
726 		struct rte_eth_dev_info *dev_info)
727 {
728 	struct pmd_internals *internals = dev->data->dev_private;
729 
730 	dev_info->if_index = internals->if_index;
731 	dev_info->max_mac_addrs = 1;
732 	dev_info->max_rx_pktlen = (uint32_t) -1;
733 	dev_info->max_rx_queues = dev->data->nb_rx_queues;
734 	dev_info->max_tx_queues = dev->data->nb_tx_queues;
735 	dev_info->min_rx_bufsize = 0;
736 
737 	return 0;
738 }
739 
740 static int
741 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
742 {
743 	unsigned int i;
744 	unsigned long rx_packets_total = 0, rx_bytes_total = 0;
745 	unsigned long rx_missed_total = 0;
746 	unsigned long tx_packets_total = 0, tx_bytes_total = 0;
747 	unsigned long tx_packets_err_total = 0;
748 	const struct pmd_internals *internal = dev->data->dev_private;
749 
750 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
751 			i < dev->data->nb_rx_queues; i++) {
752 		stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
753 		stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
754 		rx_packets_total += stats->q_ipackets[i];
755 		rx_bytes_total += stats->q_ibytes[i];
756 		rx_missed_total += queue_missed_stat_get(dev, i);
757 	}
758 
759 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
760 			i < dev->data->nb_tx_queues; i++) {
761 		stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
762 		stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
763 		tx_packets_total += stats->q_opackets[i];
764 		tx_bytes_total += stats->q_obytes[i];
765 		tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
766 	}
767 
768 	stats->ipackets = rx_packets_total;
769 	stats->ibytes = rx_bytes_total;
770 	stats->imissed = rx_missed_total;
771 	stats->opackets = tx_packets_total;
772 	stats->obytes = tx_bytes_total;
773 	stats->oerrors = tx_packets_err_total;
774 
775 	return 0;
776 }
777 
778 static int
779 eth_stats_reset(struct rte_eth_dev *dev)
780 {
781 	unsigned int i;
782 	struct pmd_internals *internal = dev->data->dev_private;
783 
784 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
785 		internal->rx_queue[i].rx_stat.pkts = 0;
786 		internal->rx_queue[i].rx_stat.bytes = 0;
787 		queue_missed_stat_reset(dev, i);
788 	}
789 
790 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
791 		internal->tx_queue[i].tx_stat.pkts = 0;
792 		internal->tx_queue[i].tx_stat.bytes = 0;
793 		internal->tx_queue[i].tx_stat.err_pkts = 0;
794 	}
795 
796 	return 0;
797 }
798 
799 static inline void
800 infinite_rx_ring_free(struct rte_ring *pkts)
801 {
802 	struct rte_mbuf *bufs;
803 
804 	while (!rte_ring_dequeue(pkts, (void **)&bufs))
805 		rte_pktmbuf_free(bufs);
806 
807 	rte_ring_free(pkts);
808 }
809 
810 static int
811 eth_dev_close(struct rte_eth_dev *dev)
812 {
813 	unsigned int i;
814 	struct pmd_internals *internals = dev->data->dev_private;
815 
816 	PMD_LOG(INFO, "Closing pcap ethdev on NUMA socket %d",
817 			rte_socket_id());
818 
819 	eth_dev_stop(dev);
820 
821 	rte_free(dev->process_private);
822 
823 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
824 		return 0;
825 
826 	/* Device wide flag, but cleanup must be performed per queue. */
827 	if (internals->infinite_rx) {
828 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
829 			struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
830 
831 			/*
832 			 * 'pcap_q->pkts' can be NULL if 'eth_dev_close()'
833 			 * called before 'eth_rx_queue_setup()' has been called
834 			 */
835 			if (pcap_q->pkts == NULL)
836 				continue;
837 
838 			infinite_rx_ring_free(pcap_q->pkts);
839 		}
840 	}
841 
842 	if (internals->phy_mac == 0)
843 		/* not dynamically allocated, must not be freed */
844 		dev->data->mac_addrs = NULL;
845 
846 	return 0;
847 }
848 
849 static void
850 eth_queue_release(void *q __rte_unused)
851 {
852 }
853 
854 static int
855 eth_link_update(struct rte_eth_dev *dev __rte_unused,
856 		int wait_to_complete __rte_unused)
857 {
858 	return 0;
859 }
860 
861 static int
862 eth_rx_queue_setup(struct rte_eth_dev *dev,
863 		uint16_t rx_queue_id,
864 		uint16_t nb_rx_desc __rte_unused,
865 		unsigned int socket_id __rte_unused,
866 		const struct rte_eth_rxconf *rx_conf __rte_unused,
867 		struct rte_mempool *mb_pool)
868 {
869 	struct pmd_internals *internals = dev->data->dev_private;
870 	struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
871 
872 	pcap_q->mb_pool = mb_pool;
873 	pcap_q->port_id = dev->data->port_id;
874 	pcap_q->queue_id = rx_queue_id;
875 	dev->data->rx_queues[rx_queue_id] = pcap_q;
876 
877 	if (internals->infinite_rx) {
878 		struct pmd_process_private *pp;
879 		char ring_name[RTE_RING_NAMESIZE];
880 		static uint32_t ring_number;
881 		uint64_t pcap_pkt_count = 0;
882 		struct rte_mbuf *bufs[1];
883 		pcap_t **pcap;
884 
885 		pp = rte_eth_devices[pcap_q->port_id].process_private;
886 		pcap = &pp->rx_pcap[pcap_q->queue_id];
887 
888 		if (unlikely(*pcap == NULL))
889 			return -ENOENT;
890 
891 		pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
892 
893 		snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu32,
894 				ring_number);
895 
896 		pcap_q->pkts = rte_ring_create(ring_name,
897 				rte_align64pow2(pcap_pkt_count + 1), 0,
898 				RING_F_SP_ENQ | RING_F_SC_DEQ);
899 		ring_number++;
900 		if (!pcap_q->pkts)
901 			return -ENOENT;
902 
903 		/* Fill ring with packets from PCAP file one by one. */
904 		while (eth_pcap_rx(pcap_q, bufs, 1)) {
905 			/* Check for multiseg mbufs. */
906 			if (bufs[0]->nb_segs != 1) {
907 				infinite_rx_ring_free(pcap_q->pkts);
908 				PMD_LOG(ERR,
909 					"Multiseg mbufs are not supported in infinite_rx mode.");
910 				return -EINVAL;
911 			}
912 
913 			rte_ring_enqueue_bulk(pcap_q->pkts,
914 					(void * const *)bufs, 1, NULL);
915 		}
916 
917 		if (rte_ring_count(pcap_q->pkts) < pcap_pkt_count) {
918 			infinite_rx_ring_free(pcap_q->pkts);
919 			PMD_LOG(ERR,
920 				"Not enough mbufs to accommodate packets in pcap file. "
921 				"At least %" PRIu64 " mbufs per queue is required.",
922 				pcap_pkt_count);
923 			return -EINVAL;
924 		}
925 
926 		/*
927 		 * Reset the stats for this queue since eth_pcap_rx calls above
928 		 * didn't result in the application receiving packets.
929 		 */
930 		pcap_q->rx_stat.pkts = 0;
931 		pcap_q->rx_stat.bytes = 0;
932 	}
933 
934 	return 0;
935 }
936 
937 static int
938 eth_tx_queue_setup(struct rte_eth_dev *dev,
939 		uint16_t tx_queue_id,
940 		uint16_t nb_tx_desc __rte_unused,
941 		unsigned int socket_id __rte_unused,
942 		const struct rte_eth_txconf *tx_conf __rte_unused)
943 {
944 	struct pmd_internals *internals = dev->data->dev_private;
945 	struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
946 
947 	pcap_q->port_id = dev->data->port_id;
948 	pcap_q->queue_id = tx_queue_id;
949 	dev->data->tx_queues[tx_queue_id] = pcap_q;
950 
951 	return 0;
952 }
953 
954 static int
955 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
956 {
957 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
958 
959 	return 0;
960 }
961 
962 static int
963 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
964 {
965 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
966 
967 	return 0;
968 }
969 
970 static int
971 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
972 {
973 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
974 
975 	return 0;
976 }
977 
978 static int
979 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
980 {
981 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
982 
983 	return 0;
984 }
985 
986 static const struct eth_dev_ops ops = {
987 	.dev_start = eth_dev_start,
988 	.dev_stop = eth_dev_stop,
989 	.dev_close = eth_dev_close,
990 	.dev_configure = eth_dev_configure,
991 	.dev_infos_get = eth_dev_info,
992 	.rx_queue_setup = eth_rx_queue_setup,
993 	.tx_queue_setup = eth_tx_queue_setup,
994 	.rx_queue_start = eth_rx_queue_start,
995 	.tx_queue_start = eth_tx_queue_start,
996 	.rx_queue_stop = eth_rx_queue_stop,
997 	.tx_queue_stop = eth_tx_queue_stop,
998 	.rx_queue_release = eth_queue_release,
999 	.tx_queue_release = eth_queue_release,
1000 	.link_update = eth_link_update,
1001 	.stats_get = eth_stats_get,
1002 	.stats_reset = eth_stats_reset,
1003 };
1004 
1005 static int
1006 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
1007 		pcap_t *pcap, pcap_dumper_t *dumper)
1008 {
1009 	if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
1010 		return -1;
1011 	if (pcap)
1012 		pmd->queue[pmd->num_of_queue].pcap = pcap;
1013 	if (dumper)
1014 		pmd->queue[pmd->num_of_queue].dumper = dumper;
1015 	pmd->queue[pmd->num_of_queue].name = name;
1016 	pmd->queue[pmd->num_of_queue].type = type;
1017 	pmd->num_of_queue++;
1018 	return 0;
1019 }
1020 
1021 /*
1022  * Function handler that opens the pcap file for reading a stores a
1023  * reference of it for use it later on.
1024  */
1025 static int
1026 open_rx_pcap(const char *key, const char *value, void *extra_args)
1027 {
1028 	const char *pcap_filename = value;
1029 	struct pmd_devargs *rx = extra_args;
1030 	pcap_t *pcap = NULL;
1031 
1032 	if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
1033 		return -1;
1034 
1035 	if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
1036 		pcap_close(pcap);
1037 		return -1;
1038 	}
1039 
1040 	return 0;
1041 }
1042 
1043 /*
1044  * Opens a pcap file for writing and stores a reference to it
1045  * for use it later on.
1046  */
1047 static int
1048 open_tx_pcap(const char *key, const char *value, void *extra_args)
1049 {
1050 	const char *pcap_filename = value;
1051 	struct pmd_devargs *dumpers = extra_args;
1052 	pcap_dumper_t *dumper;
1053 
1054 	if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
1055 		return -1;
1056 
1057 	if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
1058 		pcap_dump_close(dumper);
1059 		return -1;
1060 	}
1061 
1062 	return 0;
1063 }
1064 
1065 /*
1066  * Opens an interface for reading and writing
1067  */
1068 static inline int
1069 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
1070 {
1071 	const char *iface = value;
1072 	struct pmd_devargs *tx = extra_args;
1073 	pcap_t *pcap = NULL;
1074 
1075 	if (open_single_iface(iface, &pcap) < 0)
1076 		return -1;
1077 
1078 	tx->queue[0].pcap = pcap;
1079 	tx->queue[0].name = iface;
1080 	tx->queue[0].type = key;
1081 
1082 	return 0;
1083 }
1084 
1085 static inline int
1086 set_iface_direction(const char *iface, pcap_t *pcap,
1087 		pcap_direction_t direction)
1088 {
1089 	const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
1090 	if (pcap_setdirection(pcap, direction) < 0) {
1091 		PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1092 				iface, direction_str, pcap_geterr(pcap));
1093 		return -1;
1094 	}
1095 	PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1096 			iface, direction_str);
1097 	return 0;
1098 }
1099 
1100 static inline int
1101 open_iface(const char *key, const char *value, void *extra_args)
1102 {
1103 	const char *iface = value;
1104 	struct pmd_devargs *pmd = extra_args;
1105 	pcap_t *pcap = NULL;
1106 
1107 	if (open_single_iface(iface, &pcap) < 0)
1108 		return -1;
1109 	if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1110 		pcap_close(pcap);
1111 		return -1;
1112 	}
1113 
1114 	return 0;
1115 }
1116 
1117 /*
1118  * Opens a NIC for reading packets from it
1119  */
1120 static inline int
1121 open_rx_iface(const char *key, const char *value, void *extra_args)
1122 {
1123 	int ret = open_iface(key, value, extra_args);
1124 	if (ret < 0)
1125 		return ret;
1126 	if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1127 		struct pmd_devargs *pmd = extra_args;
1128 		unsigned int qid = pmd->num_of_queue - 1;
1129 
1130 		set_iface_direction(pmd->queue[qid].name,
1131 				pmd->queue[qid].pcap,
1132 				PCAP_D_IN);
1133 	}
1134 
1135 	return 0;
1136 }
1137 
1138 static inline int
1139 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1140 {
1141 	if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1142 			strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1143 		return open_rx_iface(key, value, extra_args);
1144 
1145 	return 0;
1146 }
1147 
1148 /*
1149  * Opens a NIC for writing packets to it
1150  */
1151 static int
1152 open_tx_iface(const char *key, const char *value, void *extra_args)
1153 {
1154 	return open_iface(key, value, extra_args);
1155 }
1156 
1157 static int
1158 select_phy_mac(const char *key __rte_unused, const char *value,
1159 		void *extra_args)
1160 {
1161 	if (extra_args) {
1162 		const int phy_mac = atoi(value);
1163 		int *enable_phy_mac = extra_args;
1164 
1165 		if (phy_mac)
1166 			*enable_phy_mac = 1;
1167 	}
1168 	return 0;
1169 }
1170 
1171 static int
1172 get_infinite_rx_arg(const char *key __rte_unused,
1173 		const char *value, void *extra_args)
1174 {
1175 	if (extra_args) {
1176 		const int infinite_rx = atoi(value);
1177 		int *enable_infinite_rx = extra_args;
1178 
1179 		if (infinite_rx > 0)
1180 			*enable_infinite_rx = 1;
1181 	}
1182 	return 0;
1183 }
1184 
1185 static int
1186 pmd_init_internals(struct rte_vdev_device *vdev,
1187 		const unsigned int nb_rx_queues,
1188 		const unsigned int nb_tx_queues,
1189 		struct pmd_internals **internals,
1190 		struct rte_eth_dev **eth_dev)
1191 {
1192 	struct rte_eth_dev_data *data;
1193 	struct pmd_process_private *pp;
1194 	unsigned int numa_node = vdev->device.numa_node;
1195 
1196 	PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1197 		numa_node);
1198 
1199 	pp = (struct pmd_process_private *)
1200 		rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1201 				RTE_CACHE_LINE_SIZE);
1202 
1203 	if (pp == NULL) {
1204 		PMD_LOG(ERR,
1205 			"Failed to allocate memory for process private");
1206 		return -1;
1207 	}
1208 
1209 	/* reserve an ethdev entry */
1210 	*eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1211 	if (!(*eth_dev)) {
1212 		rte_free(pp);
1213 		return -1;
1214 	}
1215 	(*eth_dev)->process_private = pp;
1216 	/* now put it all together
1217 	 * - store queue data in internals,
1218 	 * - store numa_node info in eth_dev
1219 	 * - point eth_dev_data to internals
1220 	 * - and point eth_dev structure to new eth_dev_data structure
1221 	 */
1222 	*internals = (*eth_dev)->data->dev_private;
1223 	/*
1224 	 * Interface MAC = 02:70:63:61:70:<iface_idx>
1225 	 * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1226 	 * where the middle 4 characters are converted to hex.
1227 	 */
1228 	(*internals)->eth_addr = (struct rte_ether_addr) {
1229 		.addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1230 	};
1231 	(*internals)->phy_mac = 0;
1232 	data = (*eth_dev)->data;
1233 	data->nb_rx_queues = (uint16_t)nb_rx_queues;
1234 	data->nb_tx_queues = (uint16_t)nb_tx_queues;
1235 	data->dev_link = pmd_link;
1236 	data->mac_addrs = &(*internals)->eth_addr;
1237 	data->promiscuous = 1;
1238 	data->all_multicast = 1;
1239 	data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1240 
1241 	/*
1242 	 * NOTE: we'll replace the data element, of originally allocated
1243 	 * eth_dev so the rings are local per-process
1244 	 */
1245 	(*eth_dev)->dev_ops = &ops;
1246 
1247 	strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1248 			ETH_PCAP_ARG_MAXLEN);
1249 
1250 	return 0;
1251 }
1252 
1253 static int
1254 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1255 		const unsigned int numa_node)
1256 {
1257 	void *mac_addrs;
1258 	struct rte_ether_addr mac;
1259 
1260 	if (osdep_iface_mac_get(if_name, &mac) < 0)
1261 		return -1;
1262 
1263 	mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1264 	if (mac_addrs == NULL)
1265 		return -1;
1266 
1267 	PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1268 	rte_memcpy(mac_addrs, mac.addr_bytes, RTE_ETHER_ADDR_LEN);
1269 	eth_dev->data->mac_addrs = mac_addrs;
1270 	return 0;
1271 }
1272 
1273 static int
1274 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1275 		struct pmd_devargs_all *devargs_all,
1276 		struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1277 {
1278 	struct pmd_process_private *pp;
1279 	struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1280 	struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1281 	const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1282 	const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1283 	unsigned int i;
1284 
1285 	if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1286 			eth_dev) < 0)
1287 		return -1;
1288 
1289 	pp = (*eth_dev)->process_private;
1290 	for (i = 0; i < nb_rx_queues; i++) {
1291 		struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1292 		struct devargs_queue *queue = &rx_queues->queue[i];
1293 
1294 		pp->rx_pcap[i] = queue->pcap;
1295 		strlcpy(rx->name, queue->name, sizeof(rx->name));
1296 		strlcpy(rx->type, queue->type, sizeof(rx->type));
1297 	}
1298 
1299 	for (i = 0; i < nb_tx_queues; i++) {
1300 		struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1301 		struct devargs_queue *queue = &tx_queues->queue[i];
1302 
1303 		pp->tx_dumper[i] = queue->dumper;
1304 		pp->tx_pcap[i] = queue->pcap;
1305 		strlcpy(tx->name, queue->name, sizeof(tx->name));
1306 		strlcpy(tx->type, queue->type, sizeof(tx->type));
1307 	}
1308 
1309 	return 0;
1310 }
1311 
1312 static int
1313 eth_from_pcaps(struct rte_vdev_device *vdev,
1314 		struct pmd_devargs_all *devargs_all)
1315 {
1316 	struct pmd_internals *internals = NULL;
1317 	struct rte_eth_dev *eth_dev = NULL;
1318 	struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1319 	int single_iface = devargs_all->single_iface;
1320 	unsigned int infinite_rx = devargs_all->infinite_rx;
1321 	int ret;
1322 
1323 	ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1324 
1325 	if (ret < 0)
1326 		return ret;
1327 
1328 	/* store weather we are using a single interface for rx/tx or not */
1329 	internals->single_iface = single_iface;
1330 
1331 	if (single_iface) {
1332 		internals->if_index =
1333 			osdep_iface_index_get(rx_queues->queue[0].name);
1334 
1335 		/* phy_mac arg is applied only only if "iface" devarg is provided */
1336 		if (rx_queues->phy_mac) {
1337 			if (eth_pcap_update_mac(rx_queues->queue[0].name,
1338 					eth_dev, vdev->device.numa_node) == 0)
1339 				internals->phy_mac = 1;
1340 		}
1341 	}
1342 
1343 	internals->infinite_rx = infinite_rx;
1344 	/* Assign rx ops. */
1345 	if (infinite_rx)
1346 		eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1347 	else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1348 			single_iface)
1349 		eth_dev->rx_pkt_burst = eth_pcap_rx;
1350 	else
1351 		eth_dev->rx_pkt_burst = eth_null_rx;
1352 
1353 	/* Assign tx ops. */
1354 	if (devargs_all->is_tx_pcap)
1355 		eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1356 	else if (devargs_all->is_tx_iface || single_iface)
1357 		eth_dev->tx_pkt_burst = eth_pcap_tx;
1358 	else
1359 		eth_dev->tx_pkt_burst = eth_tx_drop;
1360 
1361 	rte_eth_dev_probing_finish(eth_dev);
1362 	return 0;
1363 }
1364 
1365 static int
1366 pmd_pcap_probe(struct rte_vdev_device *dev)
1367 {
1368 	const char *name;
1369 	struct rte_kvargs *kvlist;
1370 	struct pmd_devargs pcaps = {0};
1371 	struct pmd_devargs dumpers = {0};
1372 	struct rte_eth_dev *eth_dev =  NULL;
1373 	struct pmd_internals *internal;
1374 	int ret = 0;
1375 
1376 	struct pmd_devargs_all devargs_all = {
1377 		.single_iface = 0,
1378 		.is_tx_pcap = 0,
1379 		.is_tx_iface = 0,
1380 		.infinite_rx = 0,
1381 	};
1382 
1383 	name = rte_vdev_device_name(dev);
1384 	PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1385 
1386 	timespec_get(&start_time, TIME_UTC);
1387 	start_cycles = rte_get_timer_cycles();
1388 	hz = rte_get_timer_hz();
1389 
1390 	ret = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
1391 			&timestamp_rx_dynflag);
1392 	if (ret != 0) {
1393 		PMD_LOG(ERR, "Failed to register Rx timestamp field/flag");
1394 		return -1;
1395 	}
1396 
1397 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1398 		eth_dev = rte_eth_dev_attach_secondary(name);
1399 		if (!eth_dev) {
1400 			PMD_LOG(ERR, "Failed to probe %s", name);
1401 			return -1;
1402 		}
1403 
1404 		internal = eth_dev->data->dev_private;
1405 
1406 		kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1407 		if (kvlist == NULL)
1408 			return -1;
1409 	} else {
1410 		kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1411 				valid_arguments);
1412 		if (kvlist == NULL)
1413 			return -1;
1414 	}
1415 
1416 	/*
1417 	 * If iface argument is passed we open the NICs and use them for
1418 	 * reading / writing
1419 	 */
1420 	if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1421 
1422 		ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1423 				&open_rx_tx_iface, &pcaps);
1424 		if (ret < 0)
1425 			goto free_kvlist;
1426 
1427 		dumpers.queue[0] = pcaps.queue[0];
1428 
1429 		ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1430 				&select_phy_mac, &pcaps.phy_mac);
1431 		if (ret < 0)
1432 			goto free_kvlist;
1433 
1434 		dumpers.phy_mac = pcaps.phy_mac;
1435 
1436 		devargs_all.single_iface = 1;
1437 		pcaps.num_of_queue = 1;
1438 		dumpers.num_of_queue = 1;
1439 
1440 		goto create_eth;
1441 	}
1442 
1443 	/*
1444 	 * We check whether we want to open a RX stream from a real NIC, a
1445 	 * pcap file or open a dummy RX stream
1446 	 */
1447 	devargs_all.is_rx_pcap =
1448 		rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1449 	devargs_all.is_rx_iface =
1450 		(rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) +
1451 		 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_IN_ARG)) ? 1 : 0;
1452 	pcaps.num_of_queue = 0;
1453 
1454 	devargs_all.is_tx_pcap =
1455 		rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1456 	devargs_all.is_tx_iface =
1457 		rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1458 	dumpers.num_of_queue = 0;
1459 
1460 	if (devargs_all.is_rx_pcap) {
1461 		/*
1462 		 * We check whether we want to infinitely rx the pcap file.
1463 		 */
1464 		unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1465 				ETH_PCAP_INFINITE_RX_ARG);
1466 
1467 		if (infinite_rx_arg_cnt == 1) {
1468 			ret = rte_kvargs_process(kvlist,
1469 					ETH_PCAP_INFINITE_RX_ARG,
1470 					&get_infinite_rx_arg,
1471 					&devargs_all.infinite_rx);
1472 			if (ret < 0)
1473 				goto free_kvlist;
1474 			PMD_LOG(INFO, "infinite_rx has been %s for %s",
1475 					devargs_all.infinite_rx ? "enabled" : "disabled",
1476 					name);
1477 
1478 		} else if (infinite_rx_arg_cnt > 1) {
1479 			PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1480 					"argument has been provided more than once "
1481 					"for %s", name);
1482 		}
1483 
1484 		ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1485 				&open_rx_pcap, &pcaps);
1486 	} else if (devargs_all.is_rx_iface) {
1487 		ret = rte_kvargs_process(kvlist, NULL,
1488 				&rx_iface_args_process, &pcaps);
1489 	} else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1490 		unsigned int i;
1491 
1492 		/* Count number of tx queue args passed before dummy rx queue
1493 		 * creation so a dummy rx queue can be created for each tx queue
1494 		 */
1495 		unsigned int num_tx_queues =
1496 			(rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1497 			rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1498 
1499 		PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1500 
1501 		/* Creating a dummy rx queue for each tx queue passed */
1502 		for (i = 0; i < num_tx_queues; i++)
1503 			ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1504 					NULL);
1505 	} else {
1506 		PMD_LOG(ERR, "Error - No rx or tx queues provided");
1507 		ret = -ENOENT;
1508 	}
1509 	if (ret < 0)
1510 		goto free_kvlist;
1511 
1512 	/*
1513 	 * We check whether we want to open a TX stream to a real NIC,
1514 	 * a pcap file, or drop packets on tx
1515 	 */
1516 	if (devargs_all.is_tx_pcap) {
1517 		ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1518 				&open_tx_pcap, &dumpers);
1519 	} else if (devargs_all.is_tx_iface) {
1520 		ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1521 				&open_tx_iface, &dumpers);
1522 	} else {
1523 		unsigned int i;
1524 
1525 		PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1526 
1527 		/* Add 1 dummy queue per rxq which counts and drops packets. */
1528 		for (i = 0; i < pcaps.num_of_queue; i++)
1529 			ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1530 					NULL);
1531 	}
1532 
1533 	if (ret < 0)
1534 		goto free_kvlist;
1535 
1536 create_eth:
1537 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1538 		struct pmd_process_private *pp;
1539 		unsigned int i;
1540 
1541 		internal = eth_dev->data->dev_private;
1542 			pp = (struct pmd_process_private *)
1543 				rte_zmalloc(NULL,
1544 					sizeof(struct pmd_process_private),
1545 					RTE_CACHE_LINE_SIZE);
1546 
1547 		if (pp == NULL) {
1548 			PMD_LOG(ERR,
1549 				"Failed to allocate memory for process private");
1550 			ret = -1;
1551 			goto free_kvlist;
1552 		}
1553 
1554 		eth_dev->dev_ops = &ops;
1555 		eth_dev->device = &dev->device;
1556 
1557 		/* setup process private */
1558 		for (i = 0; i < pcaps.num_of_queue; i++)
1559 			pp->rx_pcap[i] = pcaps.queue[i].pcap;
1560 
1561 		for (i = 0; i < dumpers.num_of_queue; i++) {
1562 			pp->tx_dumper[i] = dumpers.queue[i].dumper;
1563 			pp->tx_pcap[i] = dumpers.queue[i].pcap;
1564 		}
1565 
1566 		eth_dev->process_private = pp;
1567 		eth_dev->rx_pkt_burst = eth_pcap_rx;
1568 		if (devargs_all.is_tx_pcap)
1569 			eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1570 		else
1571 			eth_dev->tx_pkt_burst = eth_pcap_tx;
1572 
1573 		rte_eth_dev_probing_finish(eth_dev);
1574 		goto free_kvlist;
1575 	}
1576 
1577 	devargs_all.rx_queues = pcaps;
1578 	devargs_all.tx_queues = dumpers;
1579 
1580 	ret = eth_from_pcaps(dev, &devargs_all);
1581 
1582 free_kvlist:
1583 	rte_kvargs_free(kvlist);
1584 
1585 	return ret;
1586 }
1587 
1588 static int
1589 pmd_pcap_remove(struct rte_vdev_device *dev)
1590 {
1591 	struct rte_eth_dev *eth_dev = NULL;
1592 
1593 	if (!dev)
1594 		return -1;
1595 
1596 	eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1597 	if (eth_dev == NULL)
1598 		return 0; /* port already released */
1599 
1600 	eth_dev_close(eth_dev);
1601 	rte_eth_dev_release_port(eth_dev);
1602 
1603 	return 0;
1604 }
1605 
1606 static struct rte_vdev_driver pmd_pcap_drv = {
1607 	.probe = pmd_pcap_probe,
1608 	.remove = pmd_pcap_remove,
1609 };
1610 
1611 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1612 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1613 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1614 	ETH_PCAP_RX_PCAP_ARG "=<string> "
1615 	ETH_PCAP_TX_PCAP_ARG "=<string> "
1616 	ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1617 	ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1618 	ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1619 	ETH_PCAP_IFACE_ARG "=<ifc> "
1620 	ETH_PCAP_PHY_MAC_ARG "=<int>"
1621 	ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1622